1 /*
2  * Copyright (c) 2000-2002 Silicon Graphics, Inc.  All Rights Reserved.
3  *
4  * This program is free software; you can redistribute it and/or modify it
5  * under the terms of version 2 of the GNU General Public License as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it would be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
11  *
12  * Further, this software is distributed without any warranty that it is
13  * free of the rightful claim of any third person regarding infringement
14  * or the like.  Any license provided herein, whether implied or
15  * otherwise, applies only to this software file.  Patent licenses, if
16  * any, provided herein do not apply to combinations of this program with
17  * other software, or any other product whatsoever.
18  *
19  * You should have received a copy of the GNU General Public License along
20  * with this program; if not, write the Free Software Foundation, Inc., 59
21  * Temple Place - Suite 330, Boston MA 02111-1307, USA.
22  *
23  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24  * Mountain View, CA  94043, or:
25  *
26  * http://www.sgi.com
27  *
28  * For further information regarding this notice, see:
29  *
30  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
31  */
32 
33 #include "xfs.h"
34 #include "xfs_macros.h"
35 #include "xfs_types.h"
36 #include "xfs_inum.h"
37 #include "xfs_log.h"
38 #include "xfs_trans.h"
39 #include "xfs_sb.h"
40 #include "xfs_dir.h"
41 #include "xfs_dir2.h"
42 #include "xfs_dmapi.h"
43 #include "xfs_mount.h"
44 #include "xfs_ag.h"
45 #include "xfs_alloc_btree.h"
46 #include "xfs_bmap_btree.h"
47 #include "xfs_ialloc_btree.h"
48 #include "xfs_btree.h"
49 #include "xfs_attr_sf.h"
50 #include "xfs_dir_sf.h"
51 #include "xfs_dir2_sf.h"
52 #include "xfs_dinode.h"
53 #include "xfs_inode_item.h"
54 #include "xfs_inode.h"
55 #include "xfs_bmap.h"
56 #include "xfs_ialloc.h"
57 #include "xfs_itable.h"
58 #include "xfs_dfrag.h"
59 #include "xfs_error.h"
60 #include "xfs_mac.h"
61 #include "xfs_rw.h"
62 
63 /*
64  * Syssgi interface for swapext
65  */
66 int
xfs_swapext(xfs_swapext_t __user * sxp)67 xfs_swapext(
68 	xfs_swapext_t	__user *sxp)
69 {
70 	xfs_swapext_t	sx;
71 	xfs_inode_t     *ip=NULL, *tip=NULL, *ips[2];
72 	xfs_trans_t     *tp;
73 	xfs_mount_t     *mp;
74 	xfs_bstat_t	*sbp;
75 	struct file	*fp = NULL, *tfp = NULL;
76 	vnode_t		*vp, *tvp;
77 	bhv_desc_t      *bdp, *tbdp;
78 	vn_bhv_head_t   *bhp, *tbhp;
79 	uint		lock_flags=0;
80 	int		ilf_fields, tilf_fields;
81 	int		error = 0;
82 	xfs_ifork_t	tempif, *ifp, *tifp;
83 	__uint64_t	tmp;
84 	int		aforkblks = 0;
85 	int		taforkblks = 0;
86 	int		locked = 0;
87 
88 	if (copy_from_user(&sx, sxp, sizeof(sx)))
89 		return XFS_ERROR(EFAULT);
90 
91 	/* Pull information for the target fd */
92 	if (((fp = fget((int)sx.sx_fdtarget)) == NULL) ||
93 	    ((vp = LINVFS_GET_VP(fp->f_dentry->d_inode)) == NULL))  {
94 		error = XFS_ERROR(EINVAL);
95 		goto error0;
96 	}
97 
98 	bhp = VN_BHV_HEAD(vp);
99 	bdp = vn_bhv_lookup(bhp, &xfs_vnodeops);
100 	if (bdp == NULL) {
101 		error = XFS_ERROR(EBADF);
102 		goto error0;
103 	} else {
104 		ip = XFS_BHVTOI(bdp);
105 	}
106 
107 	if (((tfp = fget((int)sx.sx_fdtmp)) == NULL) ||
108 	    ((tvp = LINVFS_GET_VP(tfp->f_dentry->d_inode)) == NULL)) {
109 		error = XFS_ERROR(EINVAL);
110 		goto error0;
111 	}
112 
113 	tbhp = VN_BHV_HEAD(tvp);
114 	tbdp = vn_bhv_lookup(tbhp, &xfs_vnodeops);
115 	if (tbdp == NULL) {
116 		error = XFS_ERROR(EBADF);
117 		goto error0;
118 	} else {
119 		tip = XFS_BHVTOI(tbdp);
120 	}
121 
122 	if (ip->i_mount != tip->i_mount) {
123 		error =  XFS_ERROR(EINVAL);
124 		goto error0;
125 	}
126 
127 	if (ip->i_ino == tip->i_ino) {
128 		error =  XFS_ERROR(EINVAL);
129 		goto error0;
130 	}
131 
132 	mp = ip->i_mount;
133 
134 	sbp = &sx.sx_stat;
135 
136 	if (XFS_FORCED_SHUTDOWN(mp)) {
137 		error =  XFS_ERROR(EIO);
138 		goto error0;
139 	}
140 
141 	locked = 1;
142 
143 	/* Lock in i_ino order */
144 	if (ip->i_ino < tip->i_ino) {
145 		ips[0] = ip;
146 		ips[1] = tip;
147 	} else {
148 		ips[0] = tip;
149 		ips[1] = ip;
150 	}
151 	lock_flags = XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL;
152 	xfs_lock_inodes(ips, 2, 0, lock_flags);
153 
154 	/* Check permissions */
155 	error = xfs_iaccess(ip, S_IWUSR, NULL);
156 	if (error)
157 		goto error0;
158 
159 	error = xfs_iaccess(tip, S_IWUSR, NULL);
160 	if (error)
161 		goto error0;
162 
163 	/* Verify that both files have the same format */
164 	if ((ip->i_d.di_mode & S_IFMT) != (tip->i_d.di_mode & S_IFMT)) {
165 		error = XFS_ERROR(EINVAL);
166 		goto error0;
167 	}
168 
169 	/* Verify both files are either real-time or non-realtime */
170 	if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
171 	    (tip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
172 		error = XFS_ERROR(EINVAL);
173 		goto error0;
174 	}
175 
176 	/* Should never get a local format */
177 	if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
178 	    tip->i_d.di_format == XFS_DINODE_FMT_LOCAL) {
179 		error = XFS_ERROR(EINVAL);
180 		goto error0;
181 	}
182 
183 	if (VN_CACHED(tvp) != 0)
184 		xfs_inval_cached_pages(XFS_ITOV(tip), &(tip->i_iocore),
185 						(loff_t)0, 0, 0);
186 
187 	/* Verify O_DIRECT for ftmp */
188 	if (VN_CACHED(tvp) != 0) {
189 		error = XFS_ERROR(EINVAL);
190 		goto error0;
191 	}
192 
193 	/* Verify all data are being swapped */
194 	if (sx.sx_offset != 0 ||
195 	    sx.sx_length != ip->i_d.di_size ||
196 	    sx.sx_length != tip->i_d.di_size) {
197 		error = XFS_ERROR(EFAULT);
198 		goto error0;
199 	}
200 
201 	/*
202 	 * If the target has extended attributes, the tmp file
203 	 * must also in order to ensure the correct data fork
204 	 * format.
205 	 */
206 	if ( XFS_IFORK_Q(ip) != XFS_IFORK_Q(tip) ) {
207 		error = XFS_ERROR(EINVAL);
208 		goto error0;
209 	}
210 
211 	/*
212 	 * Compare the current change & modify times with that
213 	 * passed in.  If they differ, we abort this swap.
214 	 * This is the mechanism used to ensure the calling
215 	 * process that the file was not changed out from
216 	 * under it.
217 	 */
218 	if ((sbp->bs_ctime.tv_sec != ip->i_d.di_ctime.t_sec) ||
219 	    (sbp->bs_ctime.tv_nsec != ip->i_d.di_ctime.t_nsec) ||
220 	    (sbp->bs_mtime.tv_sec != ip->i_d.di_mtime.t_sec) ||
221 	    (sbp->bs_mtime.tv_nsec != ip->i_d.di_mtime.t_nsec)) {
222 		error = XFS_ERROR(EBUSY);
223 		goto error0;
224 	}
225 
226 	/* We need to fail if the file is memory mapped.  Once we have tossed
227 	 * all existing pages, the page fault will have no option
228 	 * but to go to the filesystem for pages. By making the page fault call
229 	 * VOP_READ (or write in the case of autogrow) they block on the iolock
230 	 * until we have switched the extents.
231 	 */
232 	if (VN_MAPPED(vp)) {
233 		error = XFS_ERROR(EBUSY);
234 		goto error0;
235 	}
236 
237 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
238 	xfs_iunlock(tip, XFS_ILOCK_EXCL);
239 
240 	/*
241 	 * There is a race condition here since we gave up the
242 	 * ilock.  However, the data fork will not change since
243 	 * we have the iolock (locked for truncation too) so we
244 	 * are safe.  We don't really care if non-io related
245 	 * fields change.
246 	 */
247 
248 	VOP_TOSS_PAGES(vp, 0, -1, FI_REMAPF);
249 
250 	tp = xfs_trans_alloc(mp, XFS_TRANS_SWAPEXT);
251 	if ((error = xfs_trans_reserve(tp, 0,
252 				     XFS_ICHANGE_LOG_RES(mp), 0,
253 				     0, 0))) {
254 		xfs_iunlock(ip,  XFS_IOLOCK_EXCL);
255 		xfs_iunlock(tip, XFS_IOLOCK_EXCL);
256 		xfs_trans_cancel(tp, 0);
257 		return error;
258 	}
259 	xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
260 
261 	/*
262 	 * Count the number of extended attribute blocks
263 	 */
264 	if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
265 	     (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
266 		error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks);
267 		if (error) {
268 			xfs_iunlock(ip,  lock_flags);
269 			xfs_iunlock(tip, lock_flags);
270 			xfs_trans_cancel(tp, 0);
271 			return error;
272 		}
273 	}
274 	if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
275 	     (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
276 		error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
277 			&taforkblks);
278 		if (error) {
279 			xfs_iunlock(ip,  lock_flags);
280 			xfs_iunlock(tip, lock_flags);
281 			xfs_trans_cancel(tp, 0);
282 			return error;
283 		}
284 	}
285 
286 	/*
287 	 * Swap the data forks of the inodes
288 	 */
289 	ifp = &ip->i_df;
290 	tifp = &tip->i_df;
291 	tempif = *ifp;	/* struct copy */
292 	*ifp = *tifp;	/* struct copy */
293 	*tifp = tempif;	/* struct copy */
294 
295 	/*
296 	 * Fix the on-disk inode values
297 	 */
298 	tmp = (__uint64_t)ip->i_d.di_nblocks;
299 	ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
300 	tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
301 
302 	tmp = (__uint64_t) ip->i_d.di_nextents;
303 	ip->i_d.di_nextents = tip->i_d.di_nextents;
304 	tip->i_d.di_nextents = tmp;
305 
306 	tmp = (__uint64_t) ip->i_d.di_format;
307 	ip->i_d.di_format = tip->i_d.di_format;
308 	tip->i_d.di_format = tmp;
309 
310 	ilf_fields = XFS_ILOG_CORE;
311 
312 	switch(ip->i_d.di_format) {
313 	case XFS_DINODE_FMT_EXTENTS:
314 		/* If the extents fit in the inode, fix the
315 		 * pointer.  Otherwise it's already NULL or
316 		 * pointing to the extent.
317 		 */
318 		if (ip->i_d.di_nextents <= XFS_INLINE_EXTS) {
319 			ifp->if_u1.if_extents =
320 				ifp->if_u2.if_inline_ext;
321 		}
322 		ilf_fields |= XFS_ILOG_DEXT;
323 		break;
324 	case XFS_DINODE_FMT_BTREE:
325 		ilf_fields |= XFS_ILOG_DBROOT;
326 		break;
327 	}
328 
329 	tilf_fields = XFS_ILOG_CORE;
330 
331 	switch(tip->i_d.di_format) {
332 	case XFS_DINODE_FMT_EXTENTS:
333 		/* If the extents fit in the inode, fix the
334 		 * pointer.  Otherwise it's already NULL or
335 		 * pointing to the extent.
336 		 */
337 		if (tip->i_d.di_nextents <= XFS_INLINE_EXTS) {
338 			tifp->if_u1.if_extents =
339 				tifp->if_u2.if_inline_ext;
340 		}
341 		tilf_fields |= XFS_ILOG_DEXT;
342 		break;
343 	case XFS_DINODE_FMT_BTREE:
344 		tilf_fields |= XFS_ILOG_DBROOT;
345 		break;
346 	}
347 
348 	/*
349 	 * Increment vnode ref counts since xfs_trans_commit &
350 	 * xfs_trans_cancel will both unlock the inodes and
351 	 * decrement the associated ref counts.
352 	 */
353 	VN_HOLD(vp);
354 	VN_HOLD(tvp);
355 
356 	xfs_trans_ijoin(tp, ip, lock_flags);
357 	xfs_trans_ijoin(tp, tip, lock_flags);
358 
359 	xfs_trans_log_inode(tp, ip,  ilf_fields);
360 	xfs_trans_log_inode(tp, tip, tilf_fields);
361 
362 	/*
363 	 * If this is a synchronous mount, make sure that the
364 	 * transaction goes to disk before returning to the user.
365 	 */
366 	if (mp->m_flags & XFS_MOUNT_WSYNC) {
367 		xfs_trans_set_sync(tp);
368 	}
369 
370 	error = xfs_trans_commit(tp, XFS_TRANS_SWAPEXT, NULL);
371 
372 	fput(fp);
373 	fput(tfp);
374 
375 	return error;
376 
377  error0:
378 	if (locked) {
379 		xfs_iunlock(ip,  lock_flags);
380 		xfs_iunlock(tip, lock_flags);
381 	}
382 
383 	if (fp != NULL) fput(fp);
384 	if (tfp != NULL) fput(tfp);
385 
386 	return error;
387 }
388